Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

[Tigerick]

As Hot Chips 34 starting this week, Intel will unveil technical information of upcoming Meteor Lake (MTL) and Arrow Lake (ARL), new generation platform after Raptor Lake. Both MTL and ARL represent new direction which Intel will move to multiple chiplets and combine as one SoC platform.

MTL also represents new compute tile that based on Intel 4 process which is based on EUV lithography, a first from Intel. Intel expects to ship MTL mobile SoC in 2023.

ARL will come after MTL so Intel should be shipping it in 2024, that is what Intel roadmap is telling us. ARL compute tile will be manufactured by Intel 20A process, a first from Intel to use GAA transistors called RibbonFET.





Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[msj10]

And for the first time, Pat showed a system running Windows 11 on a Lunar Lake CPU!

Isn't it a bit too early for an ES!

they already announced Lunar Lake tape out at the beginning of the year. they should definitely have samples for it by now.

 

[SiliconFly]

Also,

they already announced Lunar Lake tape out at the beginning of the year. they should definitely have samples for it by now.

Ya. Pat also mentioned 20A hit PDK 0.9 already.

They should be able to launch Lunar Lake well on time next year as planned.

 

[FangBLade]

Where did all that great optimism and joy go? Everything has dulled down.

 

[Geddagod]

Day 1 has been bit of a dud IMO

 

[SiliconFly]

The biggest news I think is, Pat reiterated what he already mentioned before. All nodes are healthy, on track and/or ahead of schedule!

Intel 4 is in HVM already.
Intel 3 is Manufacturing Ready already & will ramp up in Q1 2024.
Intel 20A will be Manufacturing Ready in H1 2024.
Intel 18A will follow in 6 months.

Meaning, we're gonna get ARL on time in 2024.
Followed by Lunar Lake in 6 months (Q2 2025).
Followed by Panther Lake at the end of 2025.

Thats wave after wave of CPU onslaught. Pat was pretty much excited about having Intel's execution engine back online! ✌️

 

[poke01]

Based on what?

If the efficiency claims hold up and it delivers better battery life than Raptor lake mobile

 

[eek2121]

If the efficiency claims hold up and it delivers better battery life than Raptor lake mobile

yup, I don’t care if it is the same speed as a 6+8 Raptor Lake chip, as long as it is significantly more energy efficient. Raptor Lake does a good job competing with Zen 4 in terms of absolute performance. It is well behind in terms of perf/watt. If Intel has fixed this, Meteor Lake may be in my next laptop.

 

[SiliconFly]

If the efficiency claims hold up and it delivers better battery life than Raptor lake mobile

Intel today confirmed many of the power-efficiency speculations like:

(1) CPU/GPU tiles can be shutdown independently.
(2) MTL has DLVR.
(3) SoC tile has 2 LP E-cores (TSMC N6) that consume less power than regular E-cores.
(4) New Thread Director maximizes efficiency.

Putting it all together, we can say MTL is gonna be very power efficient. But we'll never know by how much until Intel releases numbers or gives out MTL final samples.

 

[Mopetar]

Where did all that great optimism and joy go? Everything has dulled down.

What were you expecting? A live benchmarking demonstration that goes for two hours.

Intel has something that they can actually ship in three months. Even if it doesn't live up to the hype, the underlying technology is interesting and seeing how it evolves is going to be enjoyable for any technology enthusiast.

 

[Geddagod]

What were you expecting? A live benchmarking demonstration that goes for two hours.

Intel has something that they can actually ship in three months. Even if it doesn't live up to the hype, the underlying technology is interesting and seeing how it evolves is going to be enjoyable for any technology enthusiast.

No, I genuinely think this event was just a dud. A lot of people expected MTL to launch now, or at least give general perf/efficiency estimates, all we got was the GPU stuff which is still nice ig, but not really what people were looking forward too.
Roadmap updates were pretty nonexistent other than confirming PTL... exists? And their messed up slide deck showed it launching 2025, which is like lol
No die shots either (biggest L for the enthusiast community TBH)
So ye, kinda a bummer. Lot of hope that Day 2 is going to be much better : )

 

[Abwx]

What were you expecting? A live benchmarking demonstration that goes for two hours.

Intel has something that they can actually ship in three months. Even if it doesn't live up to the hype, the underlying technology is interesting and seeing how it evolves is going to be enjoyable for any technology enthusiast.

Chips wich are due for delivery in 3 monts are currently in mass producion, so at this moment it should be months that they have perfectly functional chips available for whatever demo.

 

[trivik12]

One thing confirmed is both ARL/LNL are 15th gen. I guess only u series is LNL and rest are ARL. Both of them seem to be on 20A as well since LNL was demoed today and 18A is not even in risk production. That is what Anandtech is saying at least. It would be interesting to see 2 separate chips released close to each other. Other thing about ARL is intel is saying new CPU cores and so its not Redmond Cove again.

https://imgur.com/a/iTdGUCn

 

[S'renne]

No, I genuinely think this event was just a dud. A lot of people expected MTL to launch now, or at least give general perf/efficiency estimates, all we got was the GPU stuff which is still nice ig, but not really what people were looking forward too.
Roadmap updates were pretty nonexistent other than confirming PTL... exists? And their messed up slide deck showed it launching 2025, which is like lol
No die shots either (biggest L for the enthusiast community TBH)
So ye, kinda a bummer. Lot of hope that Day 2 is going to be much better : )

Welp waiting until December for official specs/models lists

 

[S'renne]

Chips wich are due for delivery in 3 monts are currently in mass producion, so at this moment it should be months that they have perfectly functional chips available for whatever demo.

They demo'd the NPU/AI part of MTL that's about it now

 

[Tigerick]

One thing confirmed is both ARL/LNL are 15th gen. I guess only u series is LNL and rest are ARL. Both of them seem to be on 20A as well since LNL was demoed today and 18A is not even in risk production. That is what Anandtech is saying at least. It would be interesting to see 2 separate chips released close to each other. Other thing about ARL is intel is saying new CPU cores and so its not Redmond Cove again.

https://imgur.com/a/iTdGUCn

I thought the slide clearly stated LNL is either made by 18A or external node, ie TSMC's N3B node as convinced by @Exist50

 

[Mopetar]

Chips wich are due for delivery in 3 monts are currently in mass producion, so at this moment it should be months that they have perfectly functional chips available for whatever demo.

Given Intel's position there's not a lot of reason to give a performance demo, or at least nothing substantive.

If it's a dud, it just tells everyone there's nothing to get excited about. No one ever announces a dud early.

If it's really good, it just kills any current sales because people will wait three months for great performance. Given that notebooks is one of their better areas, they'd mostly be shooting themselves in the foot and upsetting their partners trying to sell current gen technology.

If it's somewhere in between then there's all the more reason to let hype build up rather than deflating any that might exist. It also lets the competition have more time to respond.

It'd be a different story if Intel only had something like 10% of the market, but given that's far from the case there's not much reason to rock the boat.

 

[Geddagod]

I thought the slide clearly stated LNL is either made by 18A or external node, ie TSMC's N3B node as convinced by @Exist50

Intel didn't say anything. Anandtech's just running with them using Intel 20A because they, for some reason, don't believe LNL is going to be fabbed externally. Their reasoning in their article was a bit weird.

 

[Geddagod]

One thing confirmed is both ARL/LNL are 15th gen. I guess only u series is LNL and rest are ARL. Both of them seem to be on 20A as well since LNL was demoed today and 18A is not even in risk production. That is what Anandtech is saying at least. It would be interesting to see 2 separate chips released close to each other. Other thing about ARL is intel is saying new CPU cores and so its not Redmond Cove again.

https://imgur.com/a/iTdGUCn

Is it just me or does it really look like LNL is using 2 chiplets in that pic? Thought it was going to be a huge monolithic die tbh

 

[S'renne]

Is it just me or does it really look like LNL is using 2 chiplets in that pic? Thought it was going to be a huge monolithic die tbh

I thought it looks like a chiplet design that's similar to Apple for maximum efficiency

 

[gdansk]

I thought it looks like a chiplet design that's similar to Apple for maximum efficiency

Hmm? Apple doesn't use chiplets especially not on the low power parts. And then they have two complete SoCs joined together on the Ultra. Chiplets are about separating functionality e.g. if the GPU was on its own die and the remainder on another die.

Edit: On second reading maybe I misunderstood. You're saying Lunar Lake is a chiplet part designed to compete with Apple's low power designs?

 

[S'renne]

Hmm? Apple doesn't use chiplets especially not on the low power parts. And then they have two complete SoCs joined together on the Ultra. Chiplets are about separating functionality e.g. if the GPU was on its own die and the remainder on another die.

Edit: On second reading maybe I misunderstood. You're saying Lunar Lake is a chiplet part designed to compete with Apple's low power designs?

Yeah it looks like it's designed to directly compete with Apple's low power designs to me

 

[Geddagod]

The biggest news I think is, Pat reiterated what he already mentioned before. All nodes are healthy, on track and/or ahead of schedule!

Intel 4 is in HVM already.
Intel 3 is Manufacturing Ready already & will ramp up in Q1 2024.
Intel 20A will be Manufacturing Ready in H1 2024.
Intel 18A will follow in 6 months.

Meaning, we're gonna get ARL on time in 2024.
Followed by Lunar Lake in 6 months (Q2 2025).
Followed by Panther Lake at the end of 2025.

Thats wave after wave of CPU onslaught. Pat was pretty much excited about having Intel's execution engine back online! ✌️

Dude, idk how many times I have to say this....
LNL launches in 2024. Intel reiterated this multiple times. There is nothing here to be confused about.

Also the wave after wave of CPU onslaught appears to be pretty mediocre. ARL looks to be marginally worse than Zen 5, and by the time PTL comes around, potentially "fixing" LNC, Zen 6 would be launching at prob around the same time.
PTL vs ARL looks like it won't be that impressive , if anything it looks very disappointing. If the rumors of PTL canning Panther Cove, the "new" architecture vs LNC, and instead using cougar cove, which is likely to just be a minor upgrade, are true, then all PTL would be is a slight node improvement, and a slight architecture improvement. At best, I would hope for some packaging changes as well (ADM cache maybe?)
I'm just hoping Nova Lake uses Royal Cove at this point. Until that product launches, nothing Intel has on desktop appears to be that exciting vs AMD.

 

[H433x0n]

Dude, idk how many times I have to say this....
LNL launches in 2024. Intel reiterated this multiple times. There is nothing here to be confused about.

Also the wave after wave of CPU onslaught appears to be pretty mediocre. ARL looks to be marginally worse than Zen 5, and by the time PTL comes around, potentially "fixing" LNC, Zen 6 would be launching at prob around the same time.
PTL vs ARL looks like it won't be that impressive , if anything it looks very disappointing. If the rumors of PTL canning Panther Cove, the "new" architecture vs LNC, and instead using cougar cove, which is likely to just be a minor upgrade, are true, then all PTL would be is a slight node improvement, and a slight architecture improvement. At best, I would hope for some packaging changes as well (ADM cache maybe?)
I'm just hoping Nova Lake uses Royal Cove at this point. Until that product launches, nothing Intel has on desktop appears to be that exciting vs AMD.

Eh, looking at the data I’m not sure if ARL will be that far behind anymore.

If it brings a +5% ST uplift over RPL-R, that’s nearly 20% better ST performance than existing Zen 4. Considering Zen 5 is a frequency regression, I think it’ll be pretty close. If Zen 5 is +/- 5% of ARL ST performance they’re sort of cooked.

Edit: To clarify what I mean - If they don’t have >5% ST performance advantage they’re not going to have significant wins anywhere. It’s not going to be more efficient, especially at idle / light load. AMD’s current chiplet configuration is less efficient than foveros. They’re going to have a pretty big bandwidth disadvantage due to using same cIOD and they’ll have a big MT deficit in the 8600X/8700X.

 

[Geddagod]

Eh, looking at the data I’m not sure if ARL will be that far behind anymore.

If it brings a +5% ST uplift over RPL-R, that’s nearly 20% better ST performance than existing Zen 4. Considering Zen 5 is a frequency regression, I think it’ll be pretty close. If Zen 5 is +/- 5% of ARL ST performance they’re sort of cooked.

Edit: To clarify what I mean - If they don’t have >5% ST performance advantage they’re not going to have significant wins anywhere. It’s not going to be more efficient, especially at idle / light load. AMD’s current chiplet configuration is less efficient than foveros. They’re going to have a pretty big bandwidth disadvantage due to using same cIOD and they’ll have a big MT deficit in the 8600X/8700X.

I don't think you can believe Zen 5 is a frequency regression and believe that IPC is going to be "only" 20%. That sounds unreasonable. There's a camp that believes IPC is going to be much higher, with an ST frequency regression by 100-200Mhz IIRC, and there's a camp that believes IPC is going to be ~20% with no frequency regression, or maybe even a slight gain.
You choosing 20% IPC increase with a frequency reduction on top just sounds like cherry picking leaks tbh.
Though I agree with a lot of your conclusions, which is why I said I think Zen 5 might only be marginally ahead, the thing is that Intel's SKU spam with ARL, LNL, and PTL coming out in rapid succession don't look to be that impressive because of what they are facing up against. They need that cycle, it looks like, to keep up with AMD.

 

[H433x0n]

I don't think you can believe Zen 5 is a frequency regression and believe that IPC is going to be "only" 20%. That sounds unreasonable. There's a camp that believes IPC is going to be much higher, with an ST frequency regression by 100-200Mhz IIRC, and there's a camp that believes IPC is going to be ~20% with no frequency regression, or maybe even a slight gain.
You choosing 20% IPC increase with a frequency reduction on top just sounds like cherry picking leaks tbh.
Though I agree with a lot of your conclusions, which is why I said I think Zen 5 might only be marginally ahead, the thing is that Intel's SKU spam with ARL, LNL, and PTL coming out in rapid succession don't look to be that impressive because of what they are facing up against. They need that cycle, it looks like, to keep up with AMD.

The people that believe it will be higher are taking it from a Turin Specint benchmark showing a 30% performance gain (as you know). I don’t think that means that Zen 5 client has a 30% IPC improvement. Milan showed a 30% improvement in Specint too and Zen 3 client launched with a 19% IPC gain.

I’ve got no data on Zen 5 client so I can’t say anything definitively but I don’t think it’s getting a 30% ST bump in a single generation. Even if it did, it’ll have 30-40% less memory bandwidth than ARL-S that it’ll be held back in certain workloads anyway.